Cardiac Arrhythmia Classification Using a Combination of Quadratic Spline-Based Wavelet Transform and Artificial Neural Classification Network.

The authors present the use of Wavelet Transform, using a quadratic spline function, and Probabilistic Neural Network (PNN) to classify 8 heartbeat conditions. The process consists of four mains stages. The first part consists of preprocessing and filtering selected ECG lead II (D II) data registers from the PhysioNet repository. The filtered signal is fed to a wavelet transform process using a quadratic spline function, to obtain a feature vector. The results are transferred to a Probabilistic Neural Network algorithm for heartbeat classification. Finally, the algorithm is tested with confusion matrices to determine classification accuracy. The algorithm yielded a 91.5%, 90.3% and 95.5% classification accuracy for auricular fibrillation, sinoauricular heart block and paroxysmal atrial fibrillation conditions respectively. The lower scores were obtained for premature atrial contraction and premature ventricular contraction conditions (75.5% and 69.9% respectively). However, considering the validation test conditions, the results suggest the algorithm is suitable for on-line classification of heartbeat conditions as part of a DSP-based Holter device.